ESTIMACIÓN DE PRECIPITACIONES MÁXIMAS, MEDIAS Y

This dissertation study compared caregiver satisfaction with hospice care between heart failure and cancer caregivers. The study had three major aims: 1) identify the predictors of family caregiver satisfaction separately for heart failure caregivers and cancer caregivers in hospice care; 2) test a model of the relationship between identified predictors and family caregiver

satisfaction with care separately in the heart failure cohort and the cancer cohort; and 3) compare family caregiver satisfaction with care between matched cohorts of hospice patients with heart failure and those with cancer. This chapter presents the methodology used to achieve these aims and is organized into six sections: a) overview of the study design, b) study sample, c) human subjects protection, d) instrumentation, e) data management and f) analytical plan.

Overview of the Study Design

A retrospective cohort design was used to achieve the aims of this study. We analyzed data from a large national hospice dataset, the National Hospice and Palliative Care Organization (NHPCO)’s National Data Set. Part of this dataset is the Family Evaluation of Hospice Care (FEHC) survey responses. The FEHC evaluates multiple domains of family caregiver satisfaction with hospice care. We used data from the 2011 FEHC survey results. Additionally, organizational data (ownership, organization size) from reporting hospices was included in the analysis. Multiple statistical methods, including multiple regression, structural equation modeling, propensity score matching and t-tests were used to analyze the data.

34

Study Sample

The targeted population was all heart failure and cancer hospice

caregivers in 2011, the latest year for which data were available. According to the National Hospice and Palliative Care Organization (NHPCO), an

estimated 1,059,000 patients died in hospice care in 2011. Of these, 11.4 percent (120,726 patients) had a primary diagnosis of heart failure while 37.7 percent (399,243 patients) had a primary diagnosis of cancer (NHPCO, 2012b).

The study sample was drawn from heart failure and cancer caregivers who were served by NHPCO member hospices in 2011 and who completed the Family Evaluation of Hospice Care (FEHC) survey after the death of their family member. NHPCO represents around 2600 hospices, about 75% of all Medicare-certified hospices nationwide (S. J. Goodlin et al., 2005; Hanson et al., 2010; NHPCO, 2013). Beginning in 2000, NHPCO began collecting yearly data on program, patient, staffing and financial statistics and also on patient and family outcomes from member hospices (Connor, Horn, Smout, &

Gassaway, 2005). In 2004, they introduced a standardized survey to measure family perceptions of hospice care that is entitled Family Evaluation of

Hospice Care (FEHC) (Connor, Teno, et al., 2005). Although this voluntary survey is sent only to NHPCO member care recipients, the demographics of past FEHC respondents are representative of the total hospice recipient population when compared to the Medicare Payment Advisory Commission (MedPAC) report released yearly (Mitchell et al., 2007).

35

NHPCO supplied a total of 90,548 FEHC responses, of which 70,765 (78.2%) were from cancer caregivers and 19,783 (21.8%) were from heart failure caregivers.

Inclusion and Exclusion Criteria

Caregivers of adult (21+ years of age) hospice patients with heart failure or cancer listed as the primary diagnosis for hospice admission who answered the FEHC in English were included in the study. Caregivers of pediatric patients, caregivers of hospice patients with another primary diagnosis and caregivers who responded to the FEHC in a non-English language were excluded. Caregivers of pediatric patients were excluded as different protocols are used in the pediatric hospice population and pediatric patients rarely die of heart failure (Organization, 2009).

Stratified Random Sampling

After selecting out those who met the inclusion/exclusion criteria, we stratified the database into heart failure caregivers and cancer caregivers. We then used a computer-generated algorithm to draw a random sample of 1000 caregivers from each stratum.

Power Analysis

Power estimation was performed to support the first aim of the study. PASS (Power Analysis and Statistical Significance) software was used to calculate the appropriate sample size for the first aim, in which multiple regression was used. The sample size of 1000 per strata achieves 90% power to detect an R2 change of 0.02 attributed to 15 independent variables

36

using an F-Test with a significance level (alpha) of 0.05 (Cohen, 1988). Please see the fifteen variables to be tested in the discussion of first aim below. It was difficult to establish R2 change based on the literature, so preliminary analyses were run to establish a baseline R2 change.

For the second aim, structural equation modeling was used. Using Jackson’s (2003) N:q rule, which states that the ratio of cases (N) to number of model parameters (q) should be ideally at least 20:1, our sample size of 1000 was more than sufficient for the number of paths analyzed (Jackson, 2003).

The third aim utilized propensity score matching. Using propensity score matching ensures that the baseline characteristics of the matched heart failure and cancer pairs will be similarly distributed (Austin, 2009). Once propensity score matching was complete, basic bivariate analyses (t-tests) were used. Because we wanted to ensure that we were able to select the best possible matches for optimal bias reduction, we opted to select matches from all respondents who met inclusion criteria and had no missing data. We ended up with 7370 matches, which was more than enough to fully power our bivariate analyses.

Propensity score matching

The end-stage cancer and end-stage heart failure populations are very different populations, in terms of demographic and clinical characteristics (Bain et al., 2009; Cheung et al.; Hauptman et al., 2007; Setoguchi et al., 2010). Our sample reflected those differences: heart failure hospice patients were on average, older, female, single and more likely to reside in a nursing home, while cancer hospice patients were, on average, younger, male,

37

married and living at home. Their caregivers were different as well – the heart failure caregivers were more likely to be older, female and the child of the patient than the cancer caregivers, who were more likely to be male and the spouse of the patient. Given how different the populations represented are, we wanted to explore whether diagnosis alone makes a difference in terms of caregiver satisfaction or if these population differences make a difference. We explored whether the population differences make a difference in Aim 1, in which we explored predictors of caregiver satisfaction. For Aim 3, we chose to utilize propensity score matching to determine if diagnosis made a difference in caregiver satisfaction, when the population differences were removed.

The propensity score represents the conditional probability that a randomly selected individual will belong to the cancer or heart failure cohort, given the observed covariates chosen (Rosenbaum, 2010). Using the

propensity score, we matched heart failure caregivers to the cancer

caregivers who most closely resembled them. The matched groups of heart failure and cancer caregivers had very similar demographic and clinical characteristics. This allowed us to examine if caregiver satisfaction varies based on diagnosis alone.

Propensity scoring does not, unfortunately, balance the two cohorts in terms of unobserved covariates. While there is no way to know how

unobserved covariates influence the outcome, it is possible to assess how great an influence an unobserved covariate would have to exert in order to significantly change the results. This was assessed via a sensitivity analysis, which was performed after the analysis was completed.

38

There are multiple methods of propensity scoring, including one-to-one, one-to-one with replacement, one-to-one with calipers, optimal matching and full propensity scoring. All of these methods aim to reduce the distance

between observations from the two cohorts and each may be the best method given defined circumstances. We estimated each method of propensity

scoring and compared the bias reduction achieved. The method that achieved the greatest bias reduction was the method used to match the two cohorts for comparison. The list of observed covariates chosen for propensity scoring and the rationale behind their selection is found under the Aim 3 analysis section below.

After propensity score matching was completed, caregiver satisfaction was compared between the two groups using t-tests. A sensitivity analysis was then performed to assess the rigor of the findings. After the entire analysis was completed, another simple random sample was drawn from each cohort and the steps of the analysis were re-run for confirmation.

Protection of Human Subjects

The data were originally collected for quality improvement and tracking purposes by NHPCO member hospices. Using the FEHC and submitting data to the NHPCO national dataset allows them to identify areas of potential improvement in practice and to compare their own results against national benchmarks and averages. There are ethical concerns to be

considered when using data originally collected for quality improvement projects rather than research. While it is well known that data derived from quality improvement projects can be utilized to study research questions and build generalizable knowledge, quality improvement faces less scrutiny and is

39

subject to less oversight than research (Nerenz, 2009). The process of FEHC administration and collection was not subject to institutional review board review, nor were the caregivers who responded to the FEHC formally

consented. This is largely because although the use of the data for research was considered possible with the creation of NHPCO’s national dataset, there were no specific research questions identified.

In designing this research study, we were mindful of the need to protect the caregivers involved. All identifying data (such as name or address) were removed by NHPCO prior to supplying the data. Furthermore, all individual hospices were identified only by a code in the dataset, rather than name. This removed the risk of an individual caregiver’s identity being revealed. The data we received from the NHPCO was not considered to meet the standards for “human subjects” according to the US Department of Health and Human Services rule 45CFR46.102(f) which defines a human subject as “a living individual about whom an investigator conducting research obtains (1) data through intervention or interaction with the individual; or (2) identifiable private information” (DHHS, 2009). The study was only conducted after undergoing expedited review and obtaining approval from the Institutional Review Board of the University of Pennsylvania.

All study data were stored in a secured file on the University of Pennsylvania School of Nursing’s server. The server was protected by a firewall and registered as a University “Critical Host” Participant. Nightly Backups and weekly backups were stored at a secure off site location. The server was monitored via the Enterprise System Monitoring Solution and has antivirus protection. All data analysis was done on a desk-top computer at the

40

School of Nursing with a password-secured user account. Instrumentation

The FEHC is a 61-item questionnaire that asks family members of hospice decedents to assess the end of life care provided (Connor, Teno, et al., 2005). The FEHC is a shortened version of the After-Death Bereaved Family Member Interview, which has been tested and used in prior research (Connor, Teno, et al., 2005) and has been endorsed by the National Quality Forum as an end of life quality care measure (Forum, 2012). Hospices

contact caregivers one to three months after the patient’s death and ask them to complete the survey. Most surveys are mailed to the caregiver and

completed with paper and pencil, but telephone administration with an established script is used by some hospices (Connor, Teno, et al., 2005). Equivalency of paper and telephone administration has been verified and documented (L. Welch, Teno, Casey, & Moorhead, 2004).

The FEHC has four domains, which examine 1) caregiver satisfaction with symptom management, 2) caregiver satisfaction with emotional support provided, 3) caregiver satisfaction with the caregiver teaching provided, and 4) coordination of care. The FEHC asks one additional question about overall family satisfaction with the hospice care provided. Appendix B contains the breakdown of which items are assigned to each domain.

Each question on the FEHC has multiple answer choices, one of which is selected to be the “desirable” answer; all others are considered “negative” answers. Scoring is done in two ways: first a problem score (the number of negative responses within a domain) is calculated and then a domain score

41

(the percentage of negative responses) is calculated. For example, domain 1 (caregiver satisfaction with symptom management) contains four items. If one negative response is given, the problem score is one, the domain score is 0.25 (25%). For both of these scores, higher numbers indicate a lower quality outcome of care. A domain score of greater than 0.20 is considered an

opportunity to improve care (Teno, Clarridge, Casey, Edgman-Levitan, & Fowler, 2001).

The FEHC also includes a question evaluating overall family

satisfaction with care. Overall satisfaction with care is measured via a five point Likert scale ranging from excellent to poor. We chose to utilize the domain scores and this one scaled question as outcomes for the analysis of aims 1-3. Utilizing domain scores allows the different domains to be more easily compared. For example, a problem score of one in the domain of symptom management, which contains eight items, is not readily comparable to a problem score of one in the domain of coordination of care, which only has three items. However, the domains scores of these two domains can be compared, as they both indicate the percentage of problems noted in that domain.

Psychometric testing of the FEHC included testing of the instrument as a whole and the individual domains. Test-retest reliability was examined via Kappa statistics for dichotomous response questions and intra-class

correlations for multi-level response questions in the original validation study (Teno et al., 2001). Kappa statistics and intra-class correlations were above 0.4 for all items, which is considered a fair to good measure of reliability (Fleiss, 1981; Teno et al., 2001). The Cronbach’s alpha was utilized as a test

42

of internal reliability for each domain and ranged from 0.58-0.87 in the initial study (Teno et al., 2001). The one domain with a Cronbach’s alpha less than 0.70, caregiver satisfaction with emotional support, was modified; subsequent testing of the current FEHC translated into Spanish yielded a Cronbach’s alpha of 0.71 (Portenoy & Teno, 2007). Teno and colleagues (2001) noted that the domains with the fewest number of items had the lowest Cronbach’s alphas, as Cronbach’s alpha is influenced by item number (Cortina, 1993).

Pearson’s or Spearmen’s correlation coefficients were used to examine inter-item and item-to-total correlations, depending on the distribution. The mean inter-item correlations for each domain ranged from 0.30-0.42 in the initial study and from 0.45-0.56 in the later study on the Spanish language version (Portenoy & Teno, 2007; Teno et al., 2001). The mean item-to-total correlations for each domain were all above 0.3 and most were roughly 0.50 in the initial study, while the mean domain item-to-total correlations ranged from 0.53-0.57 in the Spanish language version (Portenoy & Teno, 2007; Teno et al., 2001). Criterion validity (how well each problem score measures satisfaction in comparison to another measure) was measured by examining the correlation between each problem score and the 5-point scaled item on overall satisfaction. The correlation between problem scores and overall satisfaction ranged from 0.45-0.52 in the initial study (Teno et al., 2001).

The variables to be used in the analysis, with their conceptual definitions and measurement strategies are found below in Table 3.1.

Table 3.1: Variables, Definitions and Measurement Variable Name Conceptual Definition Variable type &

Measurement strategy

Sample Question Major Outcomes

43

Symptom

Management Family caregiver’s perception of the patient’s symptom severity and symptom management on the part of the hospice

Ratio: Domain score from the hospice provision of physical comfort and emotional support to the decedent domain on the FEHC

B6: How much help in dealing with his/her breathing did the patient receive while under the care of the hospice? A) less than was wanted, b) the right amount, c)more than was wanted Emotional & Spiritual Support Family caregiver’s perceptions of the emotional and spiritual support offered by the hospice, in relationship to their needs.

Ratio: Domain score from the hospice support of family emotional needs domain on the FEHC

How much emotional support did the hospice team provide to you prior to the patient’s death? A) less than was wanted, b) right amount, c) more than was wanted Caregiver teaching Caregiver’s perception of the teaching provided by the hospice on care for the patient and what to expect

Ratio: Domain score from the FEHC domain on caregiver teaching

How confident were you that you knew what to expect when the patient was dying? A)Very confident B)Somewhat confident C)Not confident Coordination of Care Caregiver’s perception of the hospice’s coordination of care for the patient

Ratio: Domain scores for the FEHC domain of coordination of care

While under the care of the hospice, was there always one nurse who was identified as being in charge of the patient’s overall care? Yes/no Overall Satisfaction Family caregiver’s perception of their overall satisfaction with the hospice care provided.

Ordinal: FEHC question G1, a scale rating of satisfaction from Poor to Excellent

Overall, how would you rate the care the patient received while under the care of the hospice? Patient and Family Demographic Variables

Age Chronological age in years

Interval: FEHC H1 (patient age) and I2 (family member’s age), which provides 19 options covering 5-year

implements from

“younger than 17” to “100 years old or older”

How old was the patient when he/she died? _____years

Gender Family perception of patient’s gender identification and family member’s own

Nominal: FEHC H2 (patient’s gender) and I3 (family member’s

gender): Male or Female

Was the patient male or female?

44

Data Management

gender identification. Family

relationship Self-identified biological or social relationship to the patient.

Nominal: FEHC question I1, categorical options: spouse, partner, child, parent, sibling, other relative, friend or other

What is your relationship to the patient?

Race/Ethnicity The ethnicity to which one most closely identifies oneself

Nominal: FEHC H5/ 6 (patient) and I5/6 (family member): five categorical race options and a Hispanic/non-Hispanic ethnicity option

Are you of Hispanic or Spanish family background?

Educational attainment

Degree status in the Western educational system

Interval: FEHC H4 (patient) and I4 (family member): 6 categorical options from less than 8th grade to more than 4- year college degree

What is the highest grade or level of school that you have completed?

Patient Clinical Variables

Length of stay Days spent in hospice care, from admission to death

Ratio: The number of days from day of

admission to day of death

For about how many days or months did the patient receive hospice services? Place of care Report of whether the

patient received care in a nursing home

Nominal dischotomous: nursing home or not

While under the care of the hospice, was the patient in a nursing home? Symptoms

experienced Family reports of physiological symptoms

experienced by the patient

Nominal: Pain, dyspnea

or depression While under the care of the hospice, did the patient have pain or take medicine for pain?

Hospice Organizational Variables Hospice agency

size The average number of patients served by the hospice annually